Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   lantianyu1986@gmail.com
To:     kys@microsoft.com, haiyangz@microsoft.com, sthemmin@microsoft.com,
        sashal@kernel.org, michael.h.kelley@microsoft.com, david@redhat.com
Cc:     Tianyu Lan <Tianyu.Lan@microsoft.com>,
        linux-hyperv@vger.kernel.org, linux-kernel@vger.kernel.org,
        vkuznets@redhat.com, eric.devolder@oracle.com
Subject: [RFC PATCH V2 5/10] x86/Hyper-V/Balloon: Avoid releasing ha_lock when traverse ha_region_list
Date:   Tue,  7 Jan 2020 21:09:45 +0800
Message-Id: <20200107130950.2983-6-Tianyu.Lan@microsoft.com>
In-Reply-To: <20200107130950.2983-1-Tianyu.Lan@microsoft.com>
References: <20200107130950.2983-1-Tianyu.Lan@microsoft.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

From: Tianyu Lan <Tianyu.Lan@microsoft.com>

ha_lock is to protect ha_region_list. It is held in
hv_online_page() and handle_pg_range(). handle_pg_range()
is to traverse ha region list, find associated hot-add region
and add memory into system. hv_online_page() is called inside
of add_memory(). Current code is to release ha_lock before
calling add_memory() to avoid holding ha_lock twice in the
hv_online_page().

To avoid releasing ha_lock, add "lock_thread" in the struct hv_
dynmem_device to record thread of traversing ha region list,
check "lock_thread" in the hv_online_page() and try holding
ha_lock when current thread is not "lock_thread".

Signed-off-by: Tianyu Lan <Tianyu.Lan@microsoft.com>
---
 drivers/hv/hv_balloon.c | 20 ++++++++++++--------
 1 file changed, 12 insertions(+), 8 deletions(-)

diff --git a/drivers/hv/hv_balloon.c b/drivers/hv/hv_balloon.c
index 185146795122..729dc5551302 100644
--- a/drivers/hv/hv_balloon.c
+++ b/drivers/hv/hv_balloon.c
@@ -547,6 +547,7 @@ struct hv_dynmem_device {
 	 * regions from ha_region_list.
 	 */
 	struct mutex ha_lock;
+	struct task_struct *lock_thread;
 
 	/*
 	 * A list of hot-add regions.
@@ -711,7 +712,6 @@ static void hv_mem_hot_add(unsigned long start, unsigned long size,
 	for (i = 0; i < (size/HA_CHUNK); i++) {
 		start_pfn = start + (i * HA_CHUNK);
 
-		mutex_lock(&dm_device.ha_lock);
 		has->ha_end_pfn +=  HA_CHUNK;
 
 		if (total_pfn > HA_CHUNK) {
@@ -723,7 +723,6 @@ static void hv_mem_hot_add(unsigned long start, unsigned long size,
 		}
 
 		has->covered_end_pfn +=  processed_pfn;
-		mutex_unlock(&dm_device.ha_lock);
 
 		init_completion(&dm_device.ol_waitevent);
 		dm_device.ha_waiting = !memhp_auto_online;
@@ -744,10 +743,8 @@ static void hv_mem_hot_add(unsigned long start, unsigned long size,
 				 */
 				do_hot_add = false;
 			}
-			mutex_lock(&dm_device.ha_lock);
 			has->ha_end_pfn -= HA_CHUNK;
 			has->covered_end_pfn -=  processed_pfn;
-			mutex_unlock(&dm_device.ha_lock);
 			break;
 		}
 
@@ -769,8 +766,14 @@ static void hv_online_page(struct page *pg, unsigned int order)
 {
 	struct hv_hotadd_state *has;
 	unsigned long pfn = page_to_pfn(pg);
+	int need_unlock;
+
+	/* If current thread hasn't hold ha_lock, take ha_lock here. */
+	if (dm_device.lock_thread != current) {
+		mutex_lock(&dm_device.ha_lock);
+		need_unlock = 1;
+	}
 
-	mutex_lock(&dm_device.ha_lock);
 	list_for_each_entry(has, &dm_device.ha_region_list, list) {
 		/* The page belongs to a different HAS. */
 		if ((pfn < has->start_pfn) ||
@@ -780,7 +783,8 @@ static void hv_online_page(struct page *pg, unsigned int order)
 		hv_bring_pgs_online(has, pfn, 1UL << order);
 		break;
 	}
-	mutex_unlock(&dm_device.ha_lock);
+	if (need_unlock)
+		mutex_unlock(&dm_device.ha_lock);
 }
 
 static int pfn_covered(unsigned long start_pfn, unsigned long pfn_cnt)
@@ -857,6 +861,7 @@ static unsigned long handle_pg_range(unsigned long pg_start,
 		pg_start);
 
 	mutex_lock(&dm_device.ha_lock);
+	dm_device.lock_thread = current;
 	list_for_each_entry(has, &dm_device.ha_region_list, list) {
 		/*
 		 * If the pfn range we are dealing with is not in the current
@@ -909,9 +914,7 @@ static unsigned long handle_pg_range(unsigned long pg_start,
 			} else {
 				pfn_cnt = size;
 			}
-			mutex_unlock(&dm_device.ha_lock);
 			hv_mem_hot_add(has->ha_end_pfn, size, pfn_cnt, has);
-			mutex_lock(&dm_device.ha_lock);
 		}
 		/*
 		 * If we managed to online any pages that were given to us,
@@ -920,6 +923,7 @@ static unsigned long handle_pg_range(unsigned long pg_start,
 		res = has->covered_end_pfn - old_covered_state;
 		break;
 	}
+	dm_device.lock_thread = NULL;
 	mutex_unlock(&dm_device.ha_lock);
 
 	return res;
-- 
2.14.5